You've been running on trust for a year. It worked, mostly. Then someone said "well, we agreed I'd own more of the product side" and it turned out you had not agreed that at all, or a lead investor sent over a diligence checklist with a line item for founder vesting schedules and you both went quiet.
This is the awkward version of the problem. Nearly every guide on how to write a cofounder agreement after starting a company is actually written for day zero, when there's no history to account for and nobody's feelings are already involved. Retrofitting is a different job. You're not agreeing on a plan, you're settling a record, and the record has two authors who remember it differently.
There are two ways to do this. They lead to different documents and different arguments. Pick deliberately.
How do you write a cofounder agreement after you've already started the company?
You choose between two approaches: Start the Clock Today, where you write a forward-looking agreement and treat the past year as settled, or Reconstruct the Year, where you first document what each person actually contributed and let that reshape the split before you sign anything. Start the Clock Today is faster and less combustible. Reconstruct the Year is more accurate and more likely to hold when someone leaves.
Both produce a real, signable agreement. The difference is what you do about the twelve months of unpaid, undocumented work sitting behind you, which is precisely the part that turns into a lawsuit later.
One note before either: you cannot backdate a document to fix tax. If you convert existing founder shares into restricted stock with a vesting schedule, the IRS 83(b) election clock runs 30 days from that new grant, not from the day you first shook hands. What you can do is give retroactive vesting credit inside a document signed today, so a founder who has been at it fourteen months starts with fourteen months of vesting already banked. That's the standard move and it's clean. Ask your lawyer to structure it that way rather than typing an old date at the top of the page.
What does the Start the Clock Today approach actually get you?
Start the Clock Today means you take the equity split you've been assuming, write it down, attach a four-year vesting schedule with a one-year cliff, credit everyone for months already served, and move on without relitigating who did more. It gets you a signed agreement in about a week, and it gets you there without either of you having to argue that your work mattered more than the other person's.
Where it holds up: when the split was genuinely uncontested and the trigger for writing things down is external. An investor asks for vesting schedules. You're hiring your first employee and need a real option pool. You're opening a bank account that wants to know who the officers are. Nobody's grievance is driving this, so you don't need to open the ledger.
It also holds up when the contribution gap is real but small. Say one of you put in fifteen hours a week for the first four months while the other was full time. That's a difference. It is not a difference worth spending a weekend on if you're now both full time and the company is working.
Where it breaks: when one of you is signing while privately thinking "this isn't right, but I'll fix it later." That thought does not go away. It surfaces eighteen months on, usually at the exact moment the company is worth something, and it surfaces as "I never really agreed to 50/50, I just signed because we needed the paperwork for the SAFE." The document you signed to prevent a fight becomes the exhibit in it.
It breaks hardest around the cliff. Retroactive credit means a founder who joined ten months ago is two months from being past the cliff and permanently owning a meaningful slice. If one of you has already half-checked-out, Start the Clock Today hands them a vested stake for the year they spent drifting. Do not use this approach if you're writing the agreement because someone's commitment is in question. You'll be papering over the exact thing the paper exists for.
What does the Reconstruct the Year approach actually get you?
Reconstruct the Year means you build a written record first: who worked what hours, who put in cash, whose code and brand and customer relationships the company is standing on, and what each of you was told the split would be. Then you adjust equity, vesting credit, or both to match, and only then do you draft. It takes a week or two longer and it produces an agreement that survives a departure.
Where it holds up: cash and time asymmetries. If one founder put $40,000 of savings into the company and the other put in none, that's not a feeling, it's a bank statement. Same with a founder who kept a salaried job for eight months. You can convert both into something concrete: treat the cash as a convertible note or as extra shares at the founding price, and give vesting credit only for months of genuine full-time work. Write the numbers next to the names.
It holds up on the pre-incorporation asset problem too. In the US, if one of you wrote the original codebase before the company existed, that code belongs to that individual until it's assigned in writing. The Copyright Act's work-for-hire provisions don't cover it, because you weren't an employee and a founding prototype isn't one of the enumerated commissioned categories. Reconstruct the Year forces that into the open, which is uncomfortable and also exactly what an acquirer's counsel will do in year four.
Where it breaks: memory. Both of you will reconstruct a year in which you were the indispensable one. Hours logged retroactively are fiction. "I was thinking about the product constantly" is not a unit of measurement and someone will try to make it one. Noam Wasserman's research in The Founder's Dilemmas (Princeton University Press, 2012) found that around 73% of founding teams split equity within a month of starting, usually with barely any discussion, and that those fast handshake splits were the ones most likely to get renegotiated. The renegotiation is the hard part, not the split.
It also breaks when the gap you uncover is bigger than the relationship can absorb. If the honest reconstruction says 70/30 and you've both been telling people it's 50/50, you now have a real fight on your hands, in the middle of a working company, with no adult in the room. Some teams should still have that fight. But go in knowing it's the likely outcome rather than being surprised by it in hour three.
Which approach survives investor diligence?
Both, if the vesting schedule is standard and the IP is assigned. Investors are not auditing whether your split is morally correct. They're checking that founders are on four-year vesting with a cliff, that all IP sits inside the company, and that nobody who left last spring is still holding 15% with no paperwork.
The thing that fails diligence is the departed contributor. Say a third person worked with you for five months, took a title, told friends they were a cofounder, and drifted off. If nothing is written, they have a colorable claim and your term sheet stalls while you find them and negotiate. Start the Clock Today doesn't touch this problem. Reconstruct the Year at least surfaces it while you still have leverage and goodwill.
So the honest answer: pick either approach for the go-forward document, but you have to reconstruct the departures regardless. That's non-negotiable.
Which clauses matter most when trust is already strained?
Four, and they're not the ones day-zero templates emphasize. When people trust each other, the equity split is the whole conversation. When trust is thin, what matters is what happens on the worst day.
1. Deadlock. "We'll talk it through" is not a mechanism. Write domains: each founder has final say inside theirs, with an explicit list of decisions that sit outside any domain and need both signatures. Something like: neither founder may unilaterally hire or terminate outside their domain, take on debt above $25,000, issue equity, or amend this agreement. If a shared decision is unresolved after 14 days, either founder may refer it to [named advisor], whose call binds both parties for 12 months. Name a real person and ask them first. An unnamed "mutually agreed third party" is a deadlock about the deadlock.
2. Departure terms. Define what happens to unvested and vested shares when someone leaves, and define "cause" narrowly and in writing (fraud, conviction, material breach, sustained failure to perform duties after written notice). Give the company a repurchase right on vested shares at fair market value for a normal exit and at original cost for a for-cause exit. Set a window: 90 days from departure. Without a window, the right expires by neglect and you find out at the worst time.
3. IP assignment, effective from the earliest contribution. Every founder signs an invention assignment covering everything built for the company, with a schedule listing prior inventions they're carving out. If one of you owns the domain personally, or the Figma files are in a personal account, or the GitHub org is under someone's name, transfer it this week. This is the clause that costs you an acquisition, not the equity split.
4. A dispute pathway one person can trigger alone. Most agreements say the founders "will attempt to resolve disputes amicably," which does nothing, because the moment you need it, one of you doesn't want to. Write it so either party can start the process unilaterally: mediation within 30 days of written notice, costs split evenly, arbitration or court only after that. Making a structured conversation the default, rather than something both people have to agree to when they're already angry, is the single highest-value sentence in the document.
How do you get everyone to sign without it feeling like an accusation?
Separate the drafting from the deciding, and bring no surprises to the table. If your cofounder first learns you think the split should change by reading a draft you wrote alone, they will read the whole document as a case against them, and they'll be right to.
The sequence that works: agree on the list of questions first, without answering any of them. Vesting start dates, what happens if someone leaves, who decides what, IP, dispute process. Send the list. Say plainly why now: "the SAFE has a founder vesting condition and we need this before the wire," or "we disagreed about the pricing call last week and I don't want the next one to be worse." A named trigger is much easier to hear than a vague desire for structure, which sounds like distrust dressed up.
Then answer the questions in the same room, in one sitting, before anyone drafts prose. Structured written agreements are also where a tool earns its place: platforms like Servanda walk cofounders through the contested points one at a time and produce a written record both people have actually responded to, which is a materially different artifact from one founder's Google Doc with comments.
One rule worth holding: no clause gets added on the day of signing. If something new appears at the end, it goes in the next amendment. Last-minute additions are how signing meetings collapse.
Which one would I pick?
Reconstruct the Year, in almost every case where the agreement is being written because of a disagreement rather than because of a form. You are writing this document precisely because the informal understanding failed. Rebuilding the same understanding with a signature under it just adds a delay to the same fight. And the reconstruction is the only version that gives you an answer when someone asks, two years out, what they were owed for that first year.
I'd switch to Start the Clock Today under three conditions. First, the trigger is purely external, an investor or a bank or a first hire, and neither founder has a grievance they'd voice under pressure. Second, the contribution gap is small enough that both of you would describe it the same way out loud. Third, you're up against a date: a wire that lands Friday, a filing deadline. In that case, sign the forward-looking agreement now with a written commitment to revisit the historical split within 60 days, and calendar it. That commitment holds only if it has a date on it. Without one, it isn't a plan, it's a way of ending the conversation.
